md: synchronize flush io with array reconfiguration

Currently rcu is used to protect iterating rdev from submit_flushes():

submit_flushes			remove_and_add_spares
				synchronize_rcu
				pers->hot_remove_disk()
 rcu_read_lock()
 rdev_for_each_rcu
  if (rdev->raid_disk >= 0)
				rdev->radi_disk = -1;
   atomic_inc(&rdev->nr_pending)
   rcu_read_unlock()
   bi = bio_alloc_bioset()
   bi->bi_end_io = md_end_flush
   bi->private = rdev
   submit_bio
   // issue io for removed rdev

Fix this problem by grabbing 'acive_io' before iterating rdev, make sure
that remove_and_add_spares() won't concurrent with submit_flushes().

Fixes: a2826aa92e2e ("md: support barrier requests on all personalities.")
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Song Liu <song@kernel.org>
Link: https://lore.kernel.org/r/20231129020234.1586910-1-yukuai1@huaweicloud.com

1 files changed, 16 insertions, 6 deletions

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 71b3397dea47..4e9fe5cbeedc 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -529,6 +529,9 @@ static void md_end_flush(struct bio *bio)
 	rdev_dec_pending(rdev, mddev);
 
 	if (atomic_dec_and_test(&mddev->flush_pending)) {
+		/* The pair is percpu_ref_get() from md_flush_request() */
+		percpu_ref_put(&mddev->active_io);
+
 		/* The pre-request flush has finished */
 		queue_work(md_wq, &mddev->flush_work);
 	}
@@ -548,12 +551,8 @@ static void submit_flushes(struct work_struct *ws)
 	rdev_for_each_rcu(rdev, mddev)
 		if (rdev->raid_disk >= 0 &&
 		    !test_bit(Faulty, &rdev->flags)) {
-			/* Take two references, one is dropped
-			 * when request finishes, one after
-			 * we reclaim rcu_read_lock
-			 */
 			struct bio *bi;
-			atomic_inc(&rdev->nr_pending);
+
 			atomic_inc(&rdev->nr_pending);
 			rcu_read_unlock();
 			bi = bio_alloc_bioset(rdev->bdev, 0,
@@ -564,7 +563,6 @@ static void submit_flushes(struct work_struct *ws)
 			atomic_inc(&mddev->flush_pending);
 			submit_bio(bi);
 			rcu_read_lock();
-			rdev_dec_pending(rdev, mddev);
 		}
 	rcu_read_unlock();
 	if (atomic_dec_and_test(&mddev->flush_pending))
@@ -617,6 +615,18 @@ bool md_flush_request(struct mddev *mddev, struct bio *bio)
 	/* new request after previous flush is completed */
 	if (ktime_after(req_start, mddev->prev_flush_start)) {
 		WARN_ON(mddev->flush_bio);
+		/*
+		 * Grab a reference to make sure mddev_suspend() will wait for
+		 * this flush to be done.
+		 *
+		 * md_flush_reqeust() is called under md_handle_request() and
+		 * 'active_io' is already grabbed, hence percpu_ref_is_zero()
+		 * won't pass, percpu_ref_tryget_live() can't be used because
+		 * percpu_ref_kill() can be called by mddev_suspend()
+		 * concurrently.
+		 */
+		WARN_ON(percpu_ref_is_zero(&mddev->active_io));
+		percpu_ref_get(&mddev->active_io);
 		mddev->flush_bio = bio;
 		bio = NULL;
 	}